Your sales history is a map of how guests really eat — not how the concept imagined they would eat on opening day. Which dishes land in the basket together, what repeats for the same guests, and how this differs between lunch and dinner: all of it shows which dish calls for which drink, which starter leads into which main, and where an apparent pairing is simply chance.
Turning that signal into a menu and recommendation rule needs filters: margin, kitchen effort, and allergens. Without them, the data may recommend something the pass cannot produce or a combination that is popular but loses money.
From a combination to a rule
Not every frequent combination deserves a prominent place. Prioritise pairings that are ordered together both more often than average and often enough not to be a coincidence — a handful of orders is not yet a pattern. For new dishes without history, start with relevant attributes — cuisine style, heat level, drink category — until enough real orders build up. Careful reasoning stops you from turning a fluke into a rule.
Think about margin, not just frequency
A combination that is ordered together often is not automatically good for the business. If the add-on is cheap or slows prep enough to trigger more remakes, it can even eat the profit. Optimise for contribution margin, not simple frequency. A slightly less common pairing with a strong margin that can be delivered quickly is often more valuable than a popular but costly classic.
Sharpen the menu itself
Sales history does more than guide recommendations; it helps you tidy up the menu. Dishes that hardly anyone orders make the menu feel more overwhelming and slow decisions down — they can go. Overloaded sections can be split so guests find what they want faster. A monthly review with the kitchen turns the data into real menu decisions. Start with clean data, though: if the website and POS use different names or prices for the same dishes, the analysis politely lies. Test orders and staff meals should be excluded too.
The 7 most common mistakes
- Building pairings from gut feeling instead of real orders.
- Treating a handful of orders as a pattern.
- Leaving new dishes without relevant attributes to guide early pairings.
- Counting frequency only, not contribution margin.
- Keeping quiet slow sellers on the menu out of habit.
- Using different names or prices on the website and POS.
- Leaving test orders and staff meals in the data.
How to make the menu data-led
Frequently asked questions
Does my sales history really know more than my experience?+
Is every frequent combination a good recommendation?+
What do I do with brand-new dishes that have no history?+
How does sales history help the menu itself?+
A menu that tells one story
Sales history is the most honest adviser your menu can have — as long as you filter its signal through margin, kitchen load, and allergens. Put real pairings where guests can see them, remove quiet losers, and work from one clean shared foundation so the menu, recommendations, and pass all tell the same story. The final piece is how a system ranks and safeguards those recommendations in real time.
